Recommendation system generation method, device, server and storage medium

Abstract

The present invention provides a method, device, server and storage medium for generating a recommendation system. The method comprises: constructing an EM model according to sample data, and iteratively training the EM model using an expected maximum algorithm to obtain maximum values ​​of position influence parameters corresponding to all position parameters; generating new sample data according to the sample data and the maximum values ​​of position influence parameters corresponding to all position parameters, training a preset estimation model according to the new sample data to obtain a new estimation model, and generating a recommendation system according to the new estimation model, thereby reducing the influence of position advantage on content recommendation and improving the matching degree between content recommended by the recommendation system and user content preference.

Description

Technical Field

[0001] The present invention relates to the field of network technology, and in particular to a recommendation system generation method, device, server and storage medium. Background Art

[0002] With the rapid development of the Internet, users can obtain a large amount of content including pictures, videos and texts from the Internet for reading and browsing. The recommendation system in the web page will recommend content to users based on their click preferences.

[0003] When users browse web pages, they pay different attention to different positions in the web pages. Content in advantageous positions on the web pages tends to receive more attention from users, that is, users are more likely to click on content in advantageous positions on the web pages. The click-through rate of users in advantageous positions on the web pages will affect the user content preference results collected by the recommendation system, resulting in a low degree of match between the content recommended by the web page recommendation system and the user content preference. In order to eliminate the influence of position advantage on content recommendation, the prior art will input web page position information as a feature into the web page recommendation system for training, so that the content recommended by the trained web page recommendation system is more in line with the user's content preference, reducing the influence of position advantage on content recommendation.

[0004] However, in the prior art, when inputting web page location information as a feature into a web page recommendation system for training, there may be a situation where the input web page location information does not match the location information actually clicked by the user. That is, the trained web page recommendation system may not be able to reduce the impact of the location information clicked by the user on the content recommendation, resulting in a problem of low matching between the content recommended by the web page recommendation system and the user's content preferences. Summary of the invention

[0005] The present invention provides a recommendation system generation method, device, server and storage medium, which improves the matching degree between the content recommended by the recommendation system and the user's content preference by constructing an expectation maximization algorithm (EM) model according to sample data to obtain the maximum values ​​of position influence parameters corresponding to all position parameters, and generating a recommendation system according to the obtained new sample data.

[0006] In a first aspect, the present invention provides a recommendation system generation method, comprising:

[0007] Acquire sample data, and construct an expectation maximization algorithm EM model according to the sample data, wherein the observable variable data of the EM model is a click operation parameter, the first latent variable of the EM model is a position influence function, the second latent variable of the EM model is a content influence function, and the expected value of the EM model is a joint probability distribution of the click operation parameter and the content influence parameter, wherein the sample data contains at least one sample parameter, each sample parameter contains a position parameter, a content parameter and a click operation parameter, the independent variable of the position influence function is a position parameter, the dependent variable of the position influence function is a position influence parameter, the independent variable of the content influence function is a content parameter, and the dependent variable of the content influence function is a content influence parameter;

[0008] Iteratively training the EM model using an expected maximum algorithm to obtain a maximum value of a position influence parameter;

[0009] Updated sample data is obtained based on the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, a preset estimation model is trained based on the updated sample data to obtain a new estimation model, and a recommendation system is generated based on the new estimation model.

[0010] In a possible design, the iterative training process of the EM model using the expectation maximization algorithm includes:

[0011] Initializing and setting position influence parameters corresponding to all position parameters in the EM model;

[0012] Obtaining an expected value of a joint probability distribution of a click operation parameter and a content influence parameter according to the position influence function, the content influence function, and the click operation parameter contained in each piece of sample data;

[0013] The maximum expected value of the position influence parameter is obtained according to the maximum likelihood estimation algorithm;

[0014] Repeat the steps of obtaining the expected value of the joint probability distribution of the click operation parameter and the content influence parameter according to the position influence function, the content influence function and the click operation parameters contained in each sample data, taking the logarithm of the expected value of the joint probability distribution of the obtained click operation parameter and the content influence parameter to obtain the expected value of the position influence parameter, and obtaining the maximum value of the expected value of the position influence parameter according to the maximum likelihood estimation algorithm, until the content influence parameters corresponding to all content parameters and the position influence parameters corresponding to all position parameters converge, and taking the obtained position influence parameters corresponding to all position parameters as the maximum values ​​of the position influence parameters corresponding to all position parameters.

[0015] In a possible design, after acquiring the sample data, the method further includes:

[0016] Obtaining historical sample data, and obtaining conversion parameters corresponding to all position parameters according to the historical sample data, wherein the conversion parameter corresponding to each position parameter is determined according to the position influence parameter corresponding to each position parameter in the pre-stored historical sample data and the click rate corresponding to each position parameter, wherein the historical sample data includes at least one historical sample parameter, and each historical sample parameter includes a position parameter, a content parameter, and a click operation parameter;

[0017] A second EM model is constructed according to the conversion parameters corresponding to all the position parameters and the sample data, wherein the observable variable data of the second EM model is the click operation parameter, the latent variable of the second EM model is the position influence parameter, and the expected value of the second EM model is the joint probability distribution of the click operation parameter and the content influence parameter;

[0018] Performing iterative training of the expected maximum algorithm on the second EM model to obtain maximum values ​​of position influence parameters corresponding to all position parameters;

[0019] Updated sample data is obtained based on the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, a preset estimation model is trained based on the updated sample data to obtain a new estimation model, and a recommendation system is generated based on the new estimation model.

[0020] In a possible design, obtaining updated sample data according to the sample data and maximum values ​​of position influence parameters corresponding to all position parameters includes:

[0021] Obtaining weight coefficients corresponding to all position parameters according to maximum values ​​of position influence parameters corresponding to all position parameters;

[0022] A new click rate corresponding to each position parameter is obtained according to the click rate corresponding to each position parameter and the weight coefficient corresponding to each position, and updated sample data is obtained according to all position parameters, the content parameters corresponding to all position parameters and the new click rates corresponding to all position parameters.

[0023] In a possible design, the weight coefficient corresponding to each position is the inverse of the maximum value of the position influence parameter corresponding to each position.

[0024] In a possible design, after generating a recommendation system according to the new estimation model, the method further includes:

[0025] sending the recommendation system to a terminal for display;

[0026] receiving new sample data sent by the terminal, where the new sample data is generated according to a click rate of a user in the recommendation system after the terminal displays the recommendation system;

[0027] According to the new sample data, repeatedly execute the steps of constructing an expected maximum algorithm EM model according to the sample data, iteratively training the EM model with the expected maximum algorithm to obtain the maximum values ​​of the position influence parameters corresponding to all position parameters, and obtaining updated sample data according to the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, training the preset estimation model according to the updated sample data to obtain a new estimation model, and generating a recommendation system according to the new estimation model.

[0028] In a second aspect, the present invention provides a recommendation system generating device, comprising:

[0029] A construction module is used to obtain sample data and construct an expectation maximization algorithm EM model according to the sample data, wherein the observable variable data of the EM model is a click operation parameter, the first latent variable of the EM model is a position influence function, the second latent variable of the EM model is a content influence function, and the expected value of the EM model is a joint probability distribution of the click operation parameter and the content influence parameter, wherein the sample data contains at least one sample parameter, each sample parameter contains a position parameter, a content parameter and a click operation parameter, the independent variable of the position influence function is a position parameter, the dependent variable of the position influence function is a position influence parameter, the independent variable of the content influence function is a content parameter, and the dependent variable of the content influence function is a content influence parameter;

[0030] A training module, used for iteratively training the EM model using an expected maximum algorithm to obtain a maximum value of a position influence parameter;

[0031] A generation module is used to obtain updated sample data based on the sample data and the maximum value of the position influence parameter corresponding to all position parameters, train a preset estimation model based on the updated sample data to obtain a new estimation model, and generate a recommendation system based on the new estimation model.

[0032] In a third aspect, the present invention provides a server, comprising: at least one processor and a memory;

[0033] The memory stores computer-executable instructions;

[0034] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the recommendation system generating method as described in the first aspect and various possible designs of the first aspect.

[0035] In a fourth aspect, the present invention provides a computer storage medium storing computer execution instructions. When a processor executes the computer execution instructions, the recommendation system generation method described in the first aspect and various possible designs of the first aspect is implemented.

[0036] In a fifth aspect, the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the recommendation system generating method as described in the first aspect and various possible designs of the first aspect.

[0037] The recommendation system generation method, device, server and storage medium provided by the present invention construct an EM model according to sample data, and iteratively train the EM model using an expected maximum algorithm to obtain the maximum values ​​of position influence parameters corresponding to all position parameters; generate new sample data according to the sample data and the maximum values ​​of position influence parameters corresponding to all position parameters, train a preset estimation model according to the new sample data to obtain a new estimation model, and generate a recommendation system according to the new estimation model, thereby reducing the influence of position advantage on content recommendation and improving the matching degree between content recommended by the recommendation system and user content preference. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0039] Figure 1 It is a schematic diagram of an application scenario of the recommendation system generation method provided by an embodiment of the present invention;

[0040] Figure 2 Schematic diagram of the process of generating a recommendation system provided by an embodiment of the present invention Figure 1 ;

[0041] Figure 3 Schematic diagram of the process of generating a recommendation system provided by an embodiment of the present invention Figure 2 ;

[0042] Figure 4 A schematic diagram of the structure of a recommendation system generating device provided by an embodiment of the present invention;

[0043] Figure 5 A schematic diagram of the hardware structure of a server provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] When users browse the web, their attention is focused on different parts of the web page, which makes the content in a "good position" get more exposure and clicks. The user's click behavior on the content in a "good position" cannot truly reflect the content quality and user preferences. In order to eliminate the impact of position advantage on content recommendation, the prior art will input the web page location information as a feature into the web page recommendation system for training, so that the content recommended by the trained web page recommendation system is more in line with the user's content preferences, reducing the impact of position advantage on content recommendation. However, in the prior art, when inputting the web page location information as a feature into the web page recommendation system for training, there may be a situation where the input web page location information does not match the location information actually clicked by the user, that is, the trained web page recommendation system may not be able to reduce the impact of the location information clicked by the user on the content recommendation, resulting in a problem of low matching between the content recommended by the web page recommendation system and the user's content preferences.

[0046] In order to solve the above technical problems, the embodiment of the present invention proposes the following technical solution: by constructing the EM model based on sample data to obtain the maximum values ​​of the position influence parameters corresponding to all position parameters, and generating a recommendation system based on the obtained new sample data, the matching degree between the content recommended by the recommendation system and the user's content preference is improved, and the influence of the position advantage on the content recommendation is reduced. The following is a detailed description using a detailed embodiment.

[0047] Figure 1 Schematic diagram of an application scenario of the recommendation system generation method provided by an embodiment of the present invention. Figure 1As shown, the terminal 101 collects the click rate, content parameters and position parameters generated during the user's web browsing process, and transmits the obtained sample data to the server 102 via a wireless network, so that the server 102 builds an EM model based on the obtained sample data, and iteratively trains the EM model with the expected maximum algorithm to obtain the maximum values ​​of the position influence parameters corresponding to all position parameters; generates new sample data based on the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, trains the preset estimation model based on the new sample data, obtains a new estimation model, and generates a recommendation system based on the new estimation model. The server 102 sends the generated recommendation system to the terminal 101, and the terminal 101 launches the recommendation system in the web browser. The recommendation system in the web page will recommend content to the user based on the user's click preference.

[0048] Figure 2 Schematic diagram of the process of generating a recommendation system provided by an embodiment of the present invention Figure 1 , the execution subject of this embodiment can be Figure 1 The server in the embodiment shown is not particularly limited in this embodiment. Figure 2 As shown, the method includes:

[0049] S201: Obtain sample data, and construct an expectation maximization algorithm EM model based on the sample data, wherein the observable variable data of the EM model is the click operation parameter, the first latent variable of the EM model is the position influence function, the second latent variable of the EM model is the content influence function, and the expected value of the EM model is the joint probability distribution of the click operation parameter and the content influence parameter, wherein the sample data contains at least one sample parameter, each sample parameter contains a position parameter, a content parameter and a click operation parameter, the independent variable of the position influence function is the position parameter, the dependent variable of the position influence function is the position influence parameter, the independent variable of the content influence function is the content parameter, and the dependent variable of the content influence function is the content influence parameter.

[0050] In an embodiment of the present invention, the web browser provided by the terminal obtains sample data corresponding to the web page according to the content data, location data and click operation contained in the user's browsing record, wherein each sample data contains the corresponding location parameter, content parameter and click operation parameter in the user's browsing record. The location parameter is the location data located on the entire page, and the content parameter is the content type. The click operation parameter is divided into two types of operation parameters: click and no click. In the background server, an EM model of the correlation relationship between implicit content parameters, location parameters and click operation parameters is preset. The EM model is obtained according to an iterative algorithm EM algorithm, and the EM model is specifically used to solve the maximum likelihood estimate of the parameters of the function model containing hidden variables.

[0051] Exemplarily, in the embodiment of the present invention, the functional relationship between the content parameter, the position parameter and the click operation parameter is as shown in formula (1):

[0052] y i =f(X i , j i ) (1)

[0053] Among them, y i represents the click operation parameters contained in the i-th sample, X i represents the content parameter of the i-th sample, j i Represents the position parameter of the i-th sample, where i is a positive integer less than m and m is the number of samples.

[0054] In the embodiment of the present invention, the functional relationship between the content parameter, the position parameter and the click operation parameter provided in formula (1) is converted into an EM model, and the position influence parameter and the content influence parameter are introduced as latent variables. In the embodiment of the present invention, the observable variable data of the EM model is the click operation parameter, the first latent variable of the EM model is the position influence function, the second latent variable of the EM model is the content influence function, and the expected value of the EM model is the joint probability distribution of the click operation parameter and the content influence parameter. Specifically, based on formula (1), the EM model constructed in the embodiment of the present invention is shown in formula (2):

[0055] y i = p(X i )*g(j i ) (2)

[0056] Among them, y i represents the click operation parameters contained in the i-th sample, X i represents the content parameter of the i-th sample, j i represents the position parameter of the i-th sample, p(x) is the relationship function between the content parameter and the content influence parameter, g(x) is the relationship function between the position parameter and the position influence parameter, p(X i ) represents the content influence parameter corresponding to the content parameter of the i-th sample, g(j i ) represents the position influence parameter corresponding to the position parameter of the i-th sample. In the embodiment of the present invention, X i In addition to the position parameter, other parameters may also be used for the i-th sample, such as device type, user type, login account type, and other parameters.

[0057] In the embodiment of the present invention, the functional relationship of p(x) is shown in formula (3):

[0058]

[0059] In the embodiment of the present invention, g(x) is a step function. i ) is g(1)=1.5, g(2)=1.3..., g(j)=0.7. Specifically, when the value of the position parameter j5 of the fifth sample is 2, the position influence parameter corresponding to the position parameter j5 of the fifth sample is 1.3.

[0060] S202: Perform iterative training of the EM model using the expectation maximization algorithm to obtain the maximum value of the position influence parameter.

[0061] In an embodiment of the present invention, illustratively, the iterative training process of the EM model using the expectation maximization algorithm includes the following steps:

[0062] S1: Initialize the position influence parameters corresponding to all position parameters in the EM model;

[0063] In the embodiment of the present invention, for the constructed expectation maximization algorithm EM model, the iteration is started by selecting the initial value of the parameter. Exemplarily, the position influence parameters corresponding to all position parameters are initialized. Specifically, the position influence parameters corresponding to all position parameters can be set to 1.0, that is, g(j i ) are all 1, where i is a positive integer.

[0064] S2: Obtain the expected value of the joint probability distribution of the click operation parameter and the content influence parameter according to the position influence function, the content influence function and the click operation parameter contained in each sample data.

[0065] In the embodiment of the present invention, for example, the joint probability distribution of the click operation parameter and the content influence parameter is set to P(y, p|g), and the logarithmic expected value corresponding to the joint probability distribution is E(log P(y, p|g)). k is the estimated value of the position influence parameter at the kth iteration, g k+1 is the estimated value of the position influence parameter at the k+1th iteration. In the k+1th iteration calculation, E(log P(y, p|g k+1 )) is affected by the click operation parameters and the position influence parameters in the previous iteration. The specific iteration formula is shown in formula (4):

[0066] Q(g k+1 , g k )=E[log P(y,p|g k+1 )|y,g k ]

[0067] =∑log P(y,p|g k+1 )P(p|y,gk ) (4)

[0068] Among them, P(p|y,g k ) is the click operation parameter y given in the sample data and the current position influence parameter g k The conditional probability distribution of the content influence parameter p in the case where k is a positive integer.

[0069] S3: Obtain the maximum expected value of the position influence parameter according to the maximum likelihood estimation algorithm.

[0070] In the embodiment of the present invention, according to all click operation parameters contained in the sample data and the position influence parameter g of the kth round k , find the value that makes Q(g k+1 , g k ) is maximized, and the estimated value of the position influence parameter of the k+1th iteration is determined.

[0071] S4: Repeat the process from S2 to S3 until the content influence parameters corresponding to all content parameters and the position influence parameters corresponding to all position parameters converge, and use the obtained position influence parameters corresponding to all position parameters as the maximum values ​​of the position influence parameters corresponding to all position parameters.

[0072] In an embodiment of the present invention, the process from S2 to S3 is repeatedly executed to perform multiple iterative calculations. When the change in the position influence parameter corresponding to each position parameter and the content influence parameter corresponding to each position parameter is less than the set change, it is considered that the two parameters meet the convergence condition, and the position influence parameters corresponding to all position parameters finally obtained are taken as the maximum values ​​of the position influence parameters corresponding to all position parameters.

[0073] S203: Obtain updated sample data based on the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, train a preset estimation model based on the updated sample data to obtain a new estimation model, and generate a recommendation system based on the new estimation model.

[0074] In an embodiment of the present invention, after obtaining the maximum values ​​of the position influence parameters corresponding to all position parameters, exemplarily, updated sample data is obtained based on the click operation parameters, content parameters, position parameters contained in the existing sample data, and the obtained maximum values ​​of the position influence parameters corresponding to all position parameters, and the updated sample data eliminates the influence of the position advantage on content recommendation by increasing the position influence parameters corresponding to all position parameters. Specifically, in an embodiment of the present invention, first, the weight coefficients corresponding to all position parameters are obtained based on the maximum values ​​of the position influence parameters corresponding to all position parameters. Specifically, the weight coefficient corresponding to each position is the inverse of the maximum value of the position influence parameter corresponding to each position. Then, new click operation parameters corresponding to each position parameter are obtained based on the click operation parameters corresponding to each position parameter and the weight coefficient corresponding to each position, and updated sample data are obtained based on all position parameters, the content parameters corresponding to all position parameters, and the new click rates corresponding to all position parameters.

[0075] In the embodiment of the present invention, the recommendation system used in the web browser is a click-through rate (CTR) prediction model. After obtaining new sample data that eliminates the influence of position advantage, the existing CTR prediction model can be trained according to the new sample data, and a recommendation system can be generated according to the prediction model obtained after training.

[0076] The recommendation system generation method provided in this embodiment uses sample data to build an EM model, and iteratively trains the EM model using an expected maximum algorithm to obtain the maximum values ​​of the position influence parameters corresponding to all position parameters, generates new sample data based on the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, trains a preset estimation model based on the new sample data to obtain a new estimation model, and generates a recommendation system based on the new estimation model, thereby reducing the impact of position advantage on content recommendation and improving the matching degree between the content recommended by the recommendation system and the user's content preference.

[0077] Figure 3 Schematic diagram of the process of generating a recommendation system provided by an embodiment of the present invention Figure 2 .exist Figure 2 Based on the examples provided, Figure 3 As shown, an embodiment of the present invention provides another method for generating a recommendation system using historical sample data, the method comprising the following steps:

[0078] S301: Acquire sample data, wherein the sample data includes at least one sample parameter, and each sample parameter includes a position parameter, a content parameter, and a click operation parameter.

[0079] S302: Obtain historical sample data, and obtain conversion parameters corresponding to all position parameters based on the historical sample data, wherein the conversion parameter corresponding to each position parameter is determined based on the position influence parameter corresponding to each position parameter in the pre-stored historical sample data and the click rate corresponding to each position parameter, wherein the historical sample data includes at least one historical sample parameter, and each historical sample parameter includes a position parameter, a content parameter, and a click operation parameter.

[0080] In an embodiment of the present invention, both the sample data and the historical sample data are data collected by the terminal from the user clicking on the web page, wherein the time period for collecting the historical sample data is after the time period for collecting the sample data, and the duration for collecting the historical sample data is greater than the duration for collecting the sample data, so as to ensure the amount of collected data contained in the collected historical sample data. After collecting the historical sample data within a preset time period, the terminal sends the collected historical sample data to the server, so that the server can obtain conversion parameters corresponding to all position parameters according to the historical sample data. Specifically, the historical sample data includes at least one historical sample parameter, and each historical sample parameter includes a position parameter, a content parameter, and a click operation parameter.

[0081] In the embodiment of the present invention, illustratively, the conversion parameter corresponding to each position parameter is determined based on the position influence parameter corresponding to each position parameter in the pre-stored historical sample data and the click rate corresponding to each position parameter. Specifically, if the position influence parameters corresponding to the same position parameter in the historical sample data are set to be the same, the content influence probability of the content corresponding to the jth position can be calculated by all the historical sample parameters contained in the historical sample data as shown in formula (5):

[0082]

[0083] Where i represents the i-th historical sample parameter in the historical sample data, y i pre represents the click rate contained in the i-th historical sample parameter in the historical sample data, g(j pre ) represents the click rate contained in the i-th historical sample parameter in the historical sample data.

[0084] S303: Construct a second EM model according to the conversion parameters corresponding to all position parameters and the sample data. The observable variable data of the second EM model is the click operation parameter, the latent variable of the second EM model is the position influence parameter, and the expected value of the second EM model is the joint probability distribution of the click operation parameter and the content influence parameter.

[0085] On the basis of the EM model, a second EM model is constructed according to the expression of the content influence probability of the content corresponding to the j-th position provided by formula (5). Specifically, the second EM model is determined according to formula (5) and formula (2) as shown in formula (6):

[0086]

[0087] S304: Perform iterative training of the expectation maximization algorithm on the second EM model to obtain maximum values ​​of position influence parameters corresponding to all position parameters.

[0088] In an embodiment of the present invention, the second EM model is iteratively trained. In the second EM model, the position influence parameter of each position is a hidden variable, that is, the second EM model is iteratively trained multiple times according to the sample data, so that the position influence parameter of each position converges and the maximum value of the position influence parameter corresponding to all position parameters is obtained.

[0089] S305: Generate new sample data based on the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, train the preset estimation model based on the new sample data to obtain a new estimation model, and generate a recommendation system based on the new estimation model.

[0090] In the embodiment of the present invention, the method of S305 is Figure 2 The method of S203 in the embodiment is consistent and will not be repeated here.

[0091] The recommendation system generation method provided in this embodiment obtains the credit score corresponding to the credit assessment probability value by using a preset scoring model, so that managers can predict the credit of the enterprise through the credit score, thereby improving the credibility of the recommendation system generation method provided in the embodiment of the present invention.

[0092] In a possible implementation, after generating a recommendation system according to a new estimation model, the recommendation system is sent to a terminal for display, and the terminal collects user web browsing data to generate new sample data, wherein the new sample data is generated according to the click rate of users in the recommendation system after the terminal displays the recommendation system. The server receives the new sample data sent by the terminal, and repeats the steps of constructing an expected maximum algorithm EM model according to the sample data, iteratively training the EM model with the expected maximum algorithm, obtaining the maximum values ​​of position influence parameters corresponding to all position parameters, generating new sample data according to the sample data and the maximum values ​​of position influence parameters corresponding to all position parameters, training a preset estimation model according to the new sample data, obtaining a new estimation model, and generating a recommendation system according to the new estimation model.

[0093] The recommendation system generation method provided in this embodiment improves the matching degree between the content recommended by the recommendation system and the user's content preference by iteratively training the EM model according to the collected new sample data after the new estimation model generates the recommendation system, obtaining a new estimation model according to the new sample data, and generating the recommendation system according to the new estimation model.

[0094] Figure 4 Schematic diagram of the structure of the recommendation system generation device provided by the embodiment of the present invention. Figure 4 As shown, the recommendation system generating device includes: a construction module 401, a training module 402 and a generating module 403.

[0095] A construction module 401 is used to obtain sample data and construct an expectation maximization algorithm EM model according to the sample data, wherein the observable variable data of the EM model is a click operation parameter, the first latent variable of the EM model is a position influence function, the second latent variable of the EM model is a content influence function, and the expected value of the EM model is a joint probability distribution of the click operation parameter and the content influence parameter, wherein the sample data contains at least one sample parameter, each sample parameter contains a position parameter, a content parameter and a click operation parameter, the independent variable of the position influence function is a position parameter, the dependent variable of the position influence function is a position influence parameter, the independent variable of the content influence function is a content parameter, and the dependent variable of the content influence function is a content influence parameter;

[0096] A training module 402 is used to perform iterative training of the EM model using an expectation maximum algorithm to obtain a maximum value of a position influence parameter;

[0097] Generation module 403 is used to obtain updated sample data based on the sample data and the maximum value of the position influence parameter corresponding to all position parameters, train the preset estimation model based on the updated sample data to obtain a new estimation model, and generate a recommendation system based on the new estimation model.

[0098] In one possible implementation, the training module 402 is specifically used to initialize the position influence parameters corresponding to all the position parameters in the EM model; obtain the expected value of the joint probability distribution of the click operation parameters and the content influence parameters according to the position influence function, the content influence function and the click operation parameters contained in each sample data; obtain the maximum value of the expected value of the content influence parameter according to the maximum likelihood estimation algorithm; repeat the steps of obtaining the expected value of the joint probability distribution of the click operation parameters and the content influence parameters according to the position influence function, the content influence function and the click operation parameters contained in each sample data, and obtaining the maximum value of the expected value of the position influence parameter according to the maximum likelihood estimation algorithm, until the content influence parameters corresponding to all the content parameters and the position influence parameters corresponding to all the position parameters converge, and use the obtained position influence parameters corresponding to all the position parameters as the maximum value of the position influence parameters corresponding to all the position parameters.

[0099] In a possible implementation, the device further includes a conversion module, which is used to obtain historical sample data, and obtain conversion parameters corresponding to all position parameters based on the historical sample data, wherein the conversion parameter corresponding to each position parameter is determined based on the position influence parameter corresponding to each position parameter in the pre-stored historical sample data and the click rate corresponding to each position parameter, wherein the historical sample data includes at least one historical sample parameter, and each historical sample parameter includes a position parameter, a content parameter, and a click operation parameter; a second EM model is constructed based on the conversion parameters corresponding to all position parameters and the sample data, wherein the observable variable data of the second EM model is the click operation parameter, the hidden variable data of the second EM model is the position influence parameter, and the expected value of the second EM model is the joint probability distribution of the click operation parameter and the content influence parameter; the second EM model is iteratively trained using an expected maximum algorithm to obtain maximum values ​​of the position influence parameters corresponding to all position parameters; updated sample data is obtained based on the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, a preset estimation model is trained based on the updated sample data to obtain a new estimation model, and a recommendation system is generated based on the new estimation model.

[0100] In one possible implementation, the generation module 403 is specifically used to obtain the weight coefficients corresponding to all position parameters based on the maximum values ​​of the position influence parameters corresponding to all the position parameters; obtain the new click rate corresponding to each position parameter based on the click rate corresponding to each position parameter and the weight coefficient corresponding to each position; and obtain updated sample data based on all position parameters, the content parameters corresponding to all position parameters, and the new click rates corresponding to all position parameters.

[0101] In a possible implementation, the device also includes a sending module, which is used to send the recommendation system to a terminal for display; receive new sample data sent by the terminal, and the new sample data is generated according to the click rate of users in the recommendation system after the terminal displays the recommendation system; repeat the steps of constructing an expected maximum algorithm EM model according to the sample data, iteratively training the EM model with the expected maximum algorithm to obtain the maximum values ​​of position influence parameters corresponding to all position parameters, and obtaining updated sample data according to the sample data and the maximum values ​​of position influence parameters corresponding to all position parameters, training a preset estimation model according to the updated sample data to obtain a new estimation model, and generating a recommendation system according to the new estimation model.

[0102] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0103] Figure 5 The hardware structure diagram of the server provided in the embodiment of the present invention is shown in FIG. Figure 5 As shown, the server of this embodiment includes: a processor 501 and a memory 502;

[0104] Memory 502, used to store computer-executable instructions;

[0105] The processor 501 is used to execute the computer-executable instructions stored in the memory to implement the various steps executed by the server in the above embodiment. For details, please refer to the relevant description in the above method embodiment.

[0106] Optionally, the memory 502 may be independent or integrated with the processor 501 .

[0107] When the memory 502 is independently provided, the server further includes a bus 503 for connecting the memory 502 and the processor 501 .

[0108] An embodiment of the present invention further provides a computer storage medium, in which computer execution instructions are stored. When a processor executes the computer execution instructions, the recommendation system generation method as described above is implemented.

[0109] The embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the recommendation system generation method as described above. The embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the recommendation system generation method as described above.

[0110] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0111] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to implement the solution of this embodiment.

[0112] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each module may exist physically separately, or two or more modules may be integrated into one unit. The above-mentioned module-composed unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0113] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the method described in each embodiment of the present application.

[0114] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor.

[0115] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0116] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0117] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0118] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0119] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a recommendation system, characterized in that: include: Acquire sample data, and construct an expectation maximization algorithm EM model according to the sample data, wherein the observable variable data of the EM model is a click operation parameter, the first latent variable of the EM model is a position influence function, the second latent variable of the EM model is a content influence function, and the expected value of the EM model is a joint probability distribution of the click operation parameter and the content influence parameter, wherein the sample data contains at least one sample parameter, each sample parameter contains a position parameter, a content parameter and a click operation parameter, the independent variable of the position influence function is a position parameter, the dependent variable of the position influence function is a position influence parameter, the independent variable of the content influence function is a content parameter, and the dependent variable of the content influence function is a content influence parameter; Iteratively training the EM model using an expected maximum algorithm to obtain a maximum value of a position influence parameter; Obtain updated sample data according to the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, train a preset estimation model according to the updated sample data to obtain a new estimation model, and generate a recommendation system according to the new estimation model; The step of obtaining updated sample data according to the sample data and maximum values ​​of position influence parameters corresponding to all position parameters includes: Obtaining weight coefficients corresponding to all position parameters according to maximum values ​​of position influence parameters corresponding to all position parameters; A new click rate corresponding to each position parameter is obtained based on the click rate corresponding to each position parameter and the weight coefficient corresponding to each position, and updated sample data is obtained based on all position parameters, the content parameters corresponding to all position parameters and the new click rates corresponding to all position parameters; the weight coefficient corresponding to each position is the inverse of the maximum value of the position influence parameter corresponding to each position.

2. The method according to claim 1, characterized in that The process of iteratively training the EM model using the expectation maximization algorithm includes: Initializing and setting position influence parameters corresponding to all position parameters in the EM model; Obtaining an expected value of a joint probability distribution of a click operation parameter and a content influence parameter according to the position influence function, the content influence function, and the click operation parameter contained in each piece of sample data; The maximum expected value of the position influence parameter is obtained according to the maximum likelihood estimation algorithm; Repeat the steps of obtaining the expected value of the joint probability distribution of the click operation parameter and the content influence parameter according to the position influence function, the content influence function and the click operation parameters contained in each sample data, taking the logarithm of the expected value of the joint probability distribution of the click operation parameter and the content influence parameter to obtain the expected value of the position influence parameter, and obtaining the maximum value of the expected value of the position influence parameter according to the maximum likelihood estimation algorithm, until the content influence parameters corresponding to all content parameters and the position influence parameters corresponding to all position parameters converge, and taking the obtained position influence parameters corresponding to all position parameters as the maximum values ​​of the position influence parameters corresponding to all position parameters.

3. The method according to claim 1, characterized in that After obtaining the sample data, the method further includes: Obtaining historical sample data, and obtaining conversion parameters corresponding to all position parameters according to the historical sample data, wherein the conversion parameter corresponding to each position parameter is determined according to the position influence parameter corresponding to each position parameter in the pre-stored historical sample data and the click rate corresponding to each position parameter, wherein the historical sample data includes at least one historical sample parameter, and each historical sample parameter includes a position parameter, a content parameter, and a click operation parameter; A second EM model is constructed according to the conversion parameters corresponding to all the position parameters and the sample data, wherein the observable variable data of the second EM model is the click operation parameter, the latent variable of the second EM model is the position influence parameter, and the expected value of the second EM model is the joint probability distribution of the click operation parameter and the content influence parameter; Performing iterative training of the expected maximum algorithm on the second EM model to obtain maximum values ​​of position influence parameters corresponding to all position parameters; Updated sample data is obtained based on the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, a preset estimation model is trained based on the updated sample data to obtain a new estimation model, and a recommendation system is generated based on the new estimation model.

4. The method according to any one of claims 1 to 3, characterized in that: After generating the recommendation system according to the new estimation model, the method further includes: sending the recommendation system to a terminal for display; receiving new sample data sent by the terminal, where the new sample data is generated according to a click rate of a user in the recommendation system after the terminal displays the recommendation system; According to the new sample data, repeatedly execute the steps of constructing an expected maximum algorithm EM model according to the sample data, iteratively training the EM model with the expected maximum algorithm to obtain the maximum values ​​of the position influence parameters corresponding to all position parameters, and obtaining updated sample data according to the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, training the preset estimation model according to the updated sample data to obtain a new estimation model, and generating a recommendation system according to the new estimation model.

5. A recommendation system generating device, characterized in that: Applicable to servers, including: A construction module is used to obtain sample data and construct an expectation maximization algorithm EM model according to the sample data, wherein the observable variable data of the EM model is a click operation parameter, the first latent variable of the EM model is a position influence function, the second latent variable of the EM model is a content influence function, and the expected value of the EM model is a joint probability distribution of the click operation parameter and the content influence parameter, wherein the sample data contains at least one sample parameter, each sample parameter contains a position parameter, a content parameter and a click operation parameter, the independent variable of the position influence function is a position parameter, the dependent variable of the position influence function is a position influence parameter, the independent variable of the content influence function is a content parameter, and the dependent variable of the content influence function is a content influence parameter; A training module, used for iteratively training the EM model using an expected maximum algorithm to obtain a maximum value of a position influence parameter; A generation module, used to obtain updated sample data according to the sample data and the maximum values ​​of the position influence parameters corresponding to all position parameters, train a preset estimation model according to the updated sample data to obtain a new estimation model, and generate a recommendation system according to the new estimation model; The generation module is specifically used to obtain weight coefficients corresponding to all position parameters according to the maximum values ​​of the position influence parameters corresponding to all position parameters; obtain a new click rate corresponding to each position parameter according to the click rate corresponding to each position parameter and the weight coefficient corresponding to each position; and obtain updated sample data according to all position parameters, content parameters corresponding to all position parameters and the new click rates corresponding to all position parameters; the weight coefficient corresponding to each position is the inverse of the maximum value of the position influence parameter corresponding to each position.

6. A server, characterized in that: include: at least one processor and memory; The memory stores computer-executable instructions; The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the recommendation system generating method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions. When the processor executes the computer-executable instructions, the recommendation system generation method according to any one of claims 1 to 4 is implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the recommendation system generating method according to any one of claims 1 to 4 is implemented.

Images